lsst.scarlet.lite.parameters.AdaproxParameter Class Reference

Inheritance diagram for lsst.scarlet.lite.parameters.AdaproxParameter:

Public Member Functions
	__init__ (self, np.ndarray x, Callable|float step, Callable|None grad=None, Callable|None prox=None, float|SingleItemArray b1=0.9, float b2=0.999, float eps=1e-8, float p=0.25, np.ndarray|None m0=None, np.ndarray|None v0=None, np.ndarray|None vhat0=None, str scheme="amsgrad", float prox_e_rel=1e-6)
	update (self, int it, np.ndarray input_grad, *args)
AdaproxParameter	__deepcopy__ (self, dict[int, Any]|None memo=None)
AdaproxParameter	__copy__ (self)
Public Member Functions inherited from lsst.scarlet.lite.parameters.Parameter
	__init__ (self, np.ndarray x, dict[str, np.ndarray] helpers, Callable|float step, Callable|None grad=None, Callable|None prox=None)
float	step (self)
tuple[int,...]	shape (self)
npt.DTypeLike	dtype (self)
Parameter	__copy__ (self)
Parameter	__deepcopy__ (self, dict[int, Any]|None memo=None)
Parameter	copy (self, bool deep=False)
	resize (self, Box old_box, Box new_box)

Public Attributes
	b1 = _b1
	b2 = b2
	eps = eps
	p = p
	scheme = scheme
	phi_psi = phi_psi[scheme]
	e_rel = prox_e_rel
Public Attributes inherited from lsst.scarlet.lite.parameters.Parameter
	x = x
	helpers = helpers
	grad = grad
	prox = prox

Additional Inherited Members
Protected Attributes inherited from lsst.scarlet.lite.parameters.Parameter
	_step = _step

Detailed Description

Operator updated using the Proximal ADAM algorithm

Uses multiple variants of adaptive quasi-Newton gradient descent

    * Adam (Kingma & Ba 2015)
    * NAdam (Dozat 2016)
    * AMSGrad (Reddi, Kale & Kumar 2018)
    * PAdam (Chen & Gu 2018)
    * AdamX (Phuong & Phong 2019)
    * RAdam (Liu et al. 2019)

See details of the algorithms in the respective papers.

Parameters
----------
x:
    The array of values that is being fit.
step:
    A numerical step value or function to calculate the step for a
    given `x`.
grad:
    A function to calculate the gradient of `x`.
prox:
    A function to take the proximal operator of `x`.
b1:
    The decay rate of the first moment (mean) of the gradient.
b2:
    The decay rate of the second moment (variance) of the gradient.
eps:
    A small constant added for numerical stability.
p:
    The power used by the ``PAdam`` scheme.
m0:
    The initial value of the first moment.
    If `None` then an array of zeros is used.
v0:
    The initial value of the second moment.
    If `None` then an array of zeros is used.
vhat0:
    The initial value of the maximum second moment.
    If `None` then an array of ``-inf`` is used.
scheme:
    The name of the ADAM variant to use to update the parameter.
prox_e_rel:
    The relative error used by the proximal operator.

Member Function Documentation

__copy__()

AdaproxParameter lsst.scarlet.lite.parameters.AdaproxParameter.__copy__

(

self

)

Create a shallow copy of this parameter.

Returns
-------
parameter:
    A shallow copy of this parameter.

__deepcopy__()

AdaproxParameter lsst.scarlet.lite.parameters.AdaproxParameter.__deepcopy__	(		self,
		dict[int, Any] | None	memo = None )

Create a deep copy of this parameter.

Parameters
----------
memo:
    A memoization dictionary used by `copy.deepcopy`.
Returns
-------
parameter:
    A deep copy of this parameter.

update()

lsst.scarlet.lite.parameters.AdaproxParameter.update	(		self,
		int	it,
		np.ndarray	input_grad,
		*	args )

Update the parameter and meta-parameters using the PGM

See `~Parameter` for more.

Reimplemented from lsst.scarlet.lite.parameters.Parameter.

---

The documentation for this class was generated from the following file:

• python/lsst/scarlet/lite/parameters.py